Adaptively weighted learning method for magnetic resonance fingerprinting
Abstract In magnetic resonance fingerprinting, every fingerprint evolution is the combined result of multiple intrinsic parameters (such as T1 and T2) and system parameters. Present learning‐based methods do not fully take into consideration of the diversity of parameters, which averages multiple pa...
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Wiley
2022-08-01
|
| Series: | IET Image Processing |
| Online Access: | https://doi.org/10.1049/ipr2.12526 |
| _version_ | 1811318944274841600 |
|---|---|
| author | Min Li Zehao Lee Zhuo Zhang |
| author_facet | Min Li Zehao Lee Zhuo Zhang |
| author_sort | Min Li |
| collection | DOAJ |
| description | Abstract In magnetic resonance fingerprinting, every fingerprint evolution is the combined result of multiple intrinsic parameters (such as T1 and T2) and system parameters. Present learning‐based methods do not fully take into consideration of the diversity of parameters, which averages multiple parameters estimation loss. Because of the non‐linear coupling nature between fingerprint evolutions and multiple parameters, different parameters have different contributions to the pattern of fingerprints. Even for the same parameter, different value ranges have different contributions to the fingerprints. During the learning processing, neglecting the diversity of parameters induces over fitting or out fitting of the network. To solve this problem, an adaptively weighted learning method is proposed. Taking the estimation uncertainty of each parameter as its weight, a weighted loss function is constructed to train the network. The weights of different parameters compete to obtain the optimal learning direction. Reconstructed fingerprints with 10% random noise is applied to train the network, and the fingerprints with different noise levels (5%–10%) are used to validate the robustness of the network. The results of simulation experiments show that the proposed method obtains better performance in terms of estimation accuracy and precision. |
| first_indexed | 2024-04-13T12:33:57Z |
| format | Article |
| id | doaj.art-d248c7cbe8c6412a8eb698ae61db6284 |
| institution | Directory Open Access Journal |
| issn | 1751-9659 1751-9667 |
| language | English |
| last_indexed | 2024-04-13T12:33:57Z |
| publishDate | 2022-08-01 |
| publisher | Wiley |
| record_format | Article |
| series | IET Image Processing |
| spelling | doaj.art-d248c7cbe8c6412a8eb698ae61db62842022-12-22T02:46:44ZengWileyIET Image Processing1751-96591751-96672022-08-0116102791280210.1049/ipr2.12526Adaptively weighted learning method for magnetic resonance fingerprintingMin Li0Zehao Lee1Zhuo Zhang2College of Internet of Things Engineering Hohai University Changzhou Jiangsu ChinaCollege of Internet of Things Engineering Hohai University Changzhou Jiangsu ChinaCollege of Internet of Things Engineering Hohai University Changzhou Jiangsu ChinaAbstract In magnetic resonance fingerprinting, every fingerprint evolution is the combined result of multiple intrinsic parameters (such as T1 and T2) and system parameters. Present learning‐based methods do not fully take into consideration of the diversity of parameters, which averages multiple parameters estimation loss. Because of the non‐linear coupling nature between fingerprint evolutions and multiple parameters, different parameters have different contributions to the pattern of fingerprints. Even for the same parameter, different value ranges have different contributions to the fingerprints. During the learning processing, neglecting the diversity of parameters induces over fitting or out fitting of the network. To solve this problem, an adaptively weighted learning method is proposed. Taking the estimation uncertainty of each parameter as its weight, a weighted loss function is constructed to train the network. The weights of different parameters compete to obtain the optimal learning direction. Reconstructed fingerprints with 10% random noise is applied to train the network, and the fingerprints with different noise levels (5%–10%) are used to validate the robustness of the network. The results of simulation experiments show that the proposed method obtains better performance in terms of estimation accuracy and precision.https://doi.org/10.1049/ipr2.12526 |
| spellingShingle | Min Li Zehao Lee Zhuo Zhang Adaptively weighted learning method for magnetic resonance fingerprinting IET Image Processing |
| title | Adaptively weighted learning method for magnetic resonance fingerprinting |
| title_full | Adaptively weighted learning method for magnetic resonance fingerprinting |
| title_fullStr | Adaptively weighted learning method for magnetic resonance fingerprinting |
| title_full_unstemmed | Adaptively weighted learning method for magnetic resonance fingerprinting |
| title_short | Adaptively weighted learning method for magnetic resonance fingerprinting |
| title_sort | adaptively weighted learning method for magnetic resonance fingerprinting |
| url | https://doi.org/10.1049/ipr2.12526 |
| work_keys_str_mv | AT minli adaptivelyweightedlearningmethodformagneticresonancefingerprinting AT zehaolee adaptivelyweightedlearningmethodformagneticresonancefingerprinting AT zhuozhang adaptivelyweightedlearningmethodformagneticresonancefingerprinting |